specVIO
// specVIO — THE PROTECTION LAYER FOR AI

Keep the Intelligence. Lose the Leak.

specVIO is a privacy layer that sits between you and every AI you use. It reads your prompt before ChatGPT, Claude, or any model does — catching API keys, emails, phone numbers, and other identifiers, and stripping them out before you hit send. Entirely local. Nothing you paste ever leaves your device.

V.I.O. — VERIFY → ISOLATE → OBFUSCATE

api_key sk-live-9a2b...→[HIDDEN_API_KEY_1]
0
BYTES TRANSMITTED
100%
ON-DEVICE
REAL-TIME
DETECTION
[ Launch Web Sandbox ]
// WHY IT MATTERS

You're not careless. The paste box just doesn't warn you.

Every one of these is a normal, harmless-feeling thing to type into a chatbot. Here's what actually goes with it — and what it looks like sanitized.

01

The Resume You Asked It to Polish

You paste your resume and cover letter into a chatbot to tighten the wording before applying.

GOES WITH IT:
Home AddressPhone NumberPast Employers

[✓] Sanitized locally — you still get the polished resume, minus the identifiers a bot never needed.

02

The Symptom You Typed at 2am

You describe a health concern to get a plain-language explanation before your appointment, name and all.

GOES WITH IT:
Full NameDate of BirthInsurance Details

[✓] Sanitized locally — you still get the explanation, without attaching it to your identity.

03

The Email Thread You Pasted for Context

You copy a work email chain in so the assistant can help you draft a reply that fits the conversation.

GOES WITH IT:
Colleagues' EmailsClient NamesInternal Deal Terms

[✓] Sanitized locally — you still get the draft, without handing over the whole thread.

These are the everyday version. For what happens at the extreme — real leaks, lawsuits, and AI malfunctions that made the news — see the incident log.

// EXPOSURE CHECK

How Exposed Are You?

Five questions about how you actually use AI chatbots. No data leaves your browser — the score is calculated on your device.

QUESTION 1 OF 5

How often do you paste real text into AI chatbots?

// PHASE 1–2 — S.P.E.C. + V.I.O.

Watch V.I.O. catch a leak, live

Paste a real prompt into the interactive sandbox and watch API keys, emails, and other identifiers get caught and redacted in your browser — nothing you type ever leaves your machine.

[ Launch Web Sandbox → ]
// CONVERSATION AUDIT

Audit Your Chat History

Not just one prompt — upload a full data export (or paste the transcript) and see everything already tied to your identity across it. The file is read in your browser tab; it's never uploaded anywhere.

[ How do I get this file? ]
CHATGPT
  1. 1. Click your profile icon → Settings
  2. 2. Data Controls → Export Data → confirm
  3. 3. Wait for the email from OpenAI (usually a few minutes)
  4. 4. Download the .zip within 24 hours — it contains conversations.json
CLAUDE
  1. 1. Click your profile icon (bottom-left) → Settings
  2. 2. Privacy → Export Data → confirm
  3. 3. Wait for the email from Anthropic (usually a few minutes)
  4. 4. Download the .zip within 24 hours — it contains conversations.json
GEMINI
  1. 1. Go to takeout.google.com and sign in
  2. 2. Deselect all → check "My Activity"
  3. 3. Open "All activity data included" → keep only "Gemini Apps"
  4. 4. Create export → wait for the email (hours to a few days)
  5. 5. Download within 7 days — contains your Gemini prompts as JSON
COPILOT
  1. 1. Go to privacy.microsoft.com and sign in
  2. 2. Privacy → Empower your productivity → Copilot
  3. 3. Your Copilot app activity history → Copilot apps
  4. 4. Export all activity history — downloads as a .csv

Microsoft doesn't offer a full JSON export yet — this CSV is the closest official option. Copy-pasting a single conversation into a .txt file works too.

GROK
  1. 1. Go to grok.com and sign in
  2. 2. Profile icon → Settings → Data Controls
  3. 3. Choose to download your data
  4. 4. Wait for it to prepare, then download the file

Other platforms usually offer something similar under Settings → Privacy or Data Controls. Once you've unzipped the download, upload or drop the conversations.json file below.

EXPORTED CONVERSATION — EDITABLE, OR DROP A FILE HERE
TOTAL LEAKS FOUND
4
Phone Numbers: 1Social Security Numbers: 1API Keys & Tokens: 1Email Addresses: 1
WHAT WE FOUND:
  • [Phone Numbers] …rah Connor, direct line (415) 555-0134 if they need to call me…
  • [Social Security Numbers] …. Here's my SSN for it: 219-09-9999. Can you format it into…
  • [API Keys & Tokens] …staging API? The key is sk-live-7f3ac91b2e4d0091. ChatGPT: Here's an exa…
  • [Email Addresses] …my personal email too — sarah.connor@resistance.org ChatGPT: Done — I've in…
[!] This conversation ties back to you in 4 places.
// VALUE PILLARS
01

Zero-Knowledge Engineering

Built on a strict local-first philosophy. Your prompts, code blocks, and documents never touch our servers.

02

The Prompt Sanitizer

Automatically scrubs API keys, connection strings, corporate names, and credentials — natively, on-device.

03

Adjustable Risk Sensitivity

Dial detection from light to strict — every keystroke is scanned live, so you always see what's at risk before you hit send.